WO2018077284A1

WO2018077284A1 - Communication method and system, electronic device and computer cluster

Info

Publication number: WO2018077284A1
Application number: PCT/CN2017/108429
Authority: WO
Inventors: 郭颖迪; 颜深根
Original assignee: 北京市商汤科技开发有限公司
Priority date: 2016-10-28
Filing date: 2017-10-30
Publication date: 2018-05-03
Also published as: US20190140982A1; US10693816B2; CN108011909B; CN108011909A

Abstract

Disclosed in the embodiments of the present application are a communication method and system, an electronic device and a computer cluster. The method comprises: establishing a corresponding thread for at least one target device among a plurality of target devices, respectively, the established thread corresponding to the target device comprising a communication thread and a message processing thread, the message processing thread comprising a message transmitting thread and/or a message receiving thread; communicating with the corresponding target device on the basis of the established corresponding thread.

Description

Communication method and system, electronic device and computer cluster

The present application claims priority to Chinese Patent Application No. 2016010967290.6, entitled "Communication Methods and Systems, Electronic Devices and Computer Clusters", filed on October 28, 2016, the entire contents of which are incorporated by reference. In the present disclosure.

Technical field

The present application relates to the field of communications technologies, and in particular, to a communication method and system, an electronic device, and a computer cluster for communicating with a device.

Background technique

In large-scale training clusters, communication is a very important module. Especially in large-scale deep learning training tasks, in order to obtain better model parameters and accelerate model convergence, frequent communication must be carried out. Communication is the bottleneck of training speed. One. At present, in the highly integrated training system and super-computing center, in order to accelerate communication, an Infiniband (IB) architecture is usually adopted. Infiniband uses dedicated hardware to simplify the protocol stack. Most of the work that requires CPU sharing between memory sharing between the two computers is done directly by IB (Infiniband) hardware.

In deep learning, communication overhead is huge, usually using Infiniband and Remote Direct Memory Access (RDMA) technology to accelerate transmission. The RDMA communication mechanism allows data to be directly transferred between the application address space and the network, bypassing the operating system kernel from the critical path of data transmission, reducing the number of memory copies, and is an efficient data transmission mechanism. In the upper training program, software that implements MPI (Message Passing Interface) is usually used as communication software, but the current MPI software can not support Infiniband well and use RDMA technology, which cannot achieve multi-thread transmission. Limits the rate at which data is transferred.

Summary of the invention

In view of this, the embodiment of the present application provides a technical solution for communicating with a device.

According to an aspect of the embodiments of the present application, a communication method is provided, including: creating a corresponding thread for at least one target device of a plurality of target devices, where the created thread corresponding to the target device includes a communication thread and a message processing thread, and the message The processing thread includes a message sending thread and/or a message receiving thread; the corresponding thread based on the configuration communicates with the corresponding target device, wherein the communication process with the first target device includes: the first message sending thread sends the information sending to the first communication thread a message, the first communication thread sends information to the first target device based on the information sending message by calling an IB interface; and/or, the first communication thread receives the information sent by the first target device by calling an IB interface. Generating an information receiving message corresponding to the received information and transmitting the message to the first message receiving thread; The first target device is one of the plurality of target devices, and the first communication thread, the first message sending thread, and the first message receiving thread are respectively the first target device Corresponding communication thread, message sending thread and message receiving thread.

According to another aspect of the present application, a communication system is provided, including: a thread configuration module, configured to create a corresponding thread for at least one target device of a plurality of target devices, where the created thread corresponding to the target device includes a communication thread and a message processing thread, the message processing thread includes a message sending thread and/or a message receiving thread; and a data communication module, configured to communicate with the corresponding target device based on the created corresponding thread; wherein the first message sending thread sends the information to the first communication thread Sending a message, the first communication thread transmitting information to the first target device based on the information sending message by calling an IB interface; and/or, the first communication thread receiving the first target device sending by calling an IB interface The information is generated by the information receiving message corresponding to the received information and sent to the first message receiving thread; the first target device is a target device of the plurality of target devices, the first communication thread, the first a message sending thread and the first message receiving thread are respectively associated with the first target device The communication thread, the thread sending the message and the message receiving thread.

According to still another aspect of the present application, an electronic device includes: a processor, a memory, an IB communication portion, and a communication bus, wherein the processor, the memory, and the communication portion complete each other through the communication bus The memory is configured to store at least one executable instruction that causes the processor to perform operations corresponding to the communication method as above.

According to still another aspect of the embodiments of the present application, a computer cluster is provided, comprising: a plurality of electronic devices as described above and a switching device respectively connected to each electronic device, and any one of the electronic devices communicates through respective IBs And communicate with other electronic devices via the switching device.

The communication method and system, the electronic device and the computer cluster of the present application adopt a multi-threading method for data transmission and reception of the target device, thereby improving the speed of data transmission and effectively utilizing the bandwidth.

The additional aspects and advantages of the embodiments of the present invention will be set forth in part in the description which follows.

DRAWINGS

The accompanying drawings, which are incorporated in FIG.

The present application can be more clearly understood from the following detailed description with reference to the accompanying drawings.

Obviously, the drawings in the following description are only some embodiments of the present invention, and those skilled in the art can obtain other drawings according to the drawings without any inventive labor:

1 is a flow chart of one embodiment of a communication method in accordance with the present application;

2 is a flowchart of processing of an information transmission message according to an embodiment of a communication method of the present application;

3 is a flow chart of processing an information receiving message according to an embodiment of a communication method of the present application;

4 is a schematic diagram of a hierarchical structure design using the present application;

5 is a block diagram of an embodiment of a communication system in accordance with the present application;

6 is a schematic diagram of one embodiment of an electronic device in accordance with the present application;

7 is a schematic diagram of one embodiment of a computer cluster in accordance with the present application.

Specific embodiment

Various exemplary embodiments of the present application will now be described in detail with reference to the drawings. It should be noted that the relative arrangement of the components and steps, numerical expressions and numerical values set forth in the embodiments are not intended to limit the scope of the application.

In the meantime, it should be understood that the dimensions of the various parts shown in the drawings are not drawn in the actual scale relationship for the convenience of the description.

The following description of the at least one exemplary embodiment is merely illustrative and is in no way

Techniques, methods and apparatus known to those of ordinary skill in the relevant art may not be discussed in detail, but the techniques, methods and apparatus should be considered as part of the specification, where appropriate.

It should be noted that similar reference numerals and letters indicate similar items in the following figures, and therefore, once an item is defined in one figure, it is not required to be further discussed in the subsequent figures.

Embodiments of the present application can be applied to computer systems/servers that can operate with numerous other general purpose or special purpose computing system environments or configurations. Examples of well-known computing systems, environments, and/or configurations suitable for use with computer systems/servers include, but are not limited to, personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, based on Microprocessor systems, set-top boxes, programmable consumer electronics, networked personal computers, small computer systems, mainframe computer systems, and distributed cloud computing technology environments including any of the above, and the like.

The computer system/server can be described in the general context of computer system executable instructions (such as program modules) being executed by a computer system. Generally, program modules may include routines, programs, target programs, components, logic, data structures, and the like that perform particular tasks or implement particular abstract data types. The computer system/server can be implemented in a distributed cloud computing environment where tasks are performed by remote processing devices that are linked through a communication network. In a distributed cloud computing environment, program modules may be located on a local or remote computing system storage medium including storage devices.

The "first" below is used to describe the upper phase difference and has no other special meaning.

The communication method of the present application adopts multi-threading technology for data interaction of multiple target devices, and can adopt multi-threading technology only for data transmission or reception, and multi-threading technology for data transmission and reception. Corresponding threads are respectively created for at least one or each of the plurality of target devices, for example, a communication thread and a message processing thread corresponding to any target device are created, and the message processing thread includes a message sending thread and a message receiving thread. For example, the first target device is any one of the plurality of target devices, and the first communication thread, the first message sending thread, and the first message receiving thread are respectively a communication thread corresponding to the first target device, a message sending thread, and The message receiving thread.

1 is a flow chart according to an embodiment of a communication method, as shown in FIG. 1:

Step 101: The first message sending thread sends an information sending message to the first communications thread. For example, the information sending message may be information about a user purchasing a product sent to a target server.

Step 102: The first communication thread sends information to the first target device based on the information sending message by calling the IB interface. The information transmitted to and received from the target device in the present application includes data and/or control commands.

Step 103: The first communication thread receives the data sent by the first target device by calling the IB interface, generates an information receiving message corresponding to the received data, and sends the information to the first message receiving thread.

The first communication thread processes the received information sending message, for example, extracts the product information purchased by the user from the information sending message, and generates a sending data instruction according to a preset rule, and transmits the sending data instruction to the IB interface, and invokes the IB interface. The function is sent to the target device.

Data can be sent between threads using global variables, message mechanisms, and so on. The information sending message may be in multiple formats, including a message header, a message body, etc., and the corresponding target device information, the message sending thread, and the communication thread information are encapsulated in the message header, and the data to be sent is encapsulated in the message body. The information receiving message may also be in multiple formats, including a message header, a message body, etc., and the corresponding target device information, the message receiving thread, and the communication thread information are encapsulated in the message header, and the received data sent by the target device is encapsulated in the message body. .

It is determined whether the data is sent to the first target device for the first time, and if so, the first message sending thread corresponding to the first target device is created. If no, the existing first message sending thread is used. It is determined whether the information sent by the first target device is received for the first time, and if so, the first message receiving thread is created, and if not, the existing first message receiving thread is used.

A plurality of target devices are created with respective communication threads for receiving and transmitting information, and receiving and transmitting various control commands, feedback information, and the like. The IB interface may include, but is not limited to, IB VERBS, SDP (Sockets Direct Protocol), and IPoIB (Internet Protocol over InfiniBand). For example, by directly calling the IB VERBS interface or driver, a control command can be directly sent to the IB VERBS interface or driver during IB transmission, and information can be transmitted using the RDMA method.

Determining whether the first communication thread corresponding to the first target device has been created, and if so, directly using the first The communication thread transmits and receives information, and if not, creates a first communication thread. Each communication thread is assigned its rank number (identification number) when it creates the communication thread, and the rank number is used to identify the communication thread. For example, the first communication thread establishes a connection with the first target device through the socket by calling the IB interface, and exchanges the RDMA communication environment information, such as the buffer address and the quantity, through the established socket connection, and the information transmission is performed through the RDMA mechanism. The first communication thread receives feedback information of the first target device, for example, data transmission success, failure, etc., and delivers the feedback information to the first message sending thread.

The first communication thread receives the information sent by the first target device by calling the IB interface, and the first communication thread delivers the information receiving message corresponding to the first target device to the first message receiving thread corresponding to the first target device, and the first message is received. The thread processes the information receiving message, for example, the first message receiving thread extracts data from the information receiving message, and writes the data to a database or the like. After processing the information receiving message, the first message receiving thread transmits the feedback information to the first communication thread, and the first communication thread sends the message to the target device.

In the communication method in the above embodiment, a corresponding communication thread is established for the target device, and a message sending thread and a message receiving thread are used, and the communication thread is used to invoke the IB interface for data communication, and the message sending thread and the message receiving thread can implement the sending. And the asynchronous operation of receiving information can improve the efficiency of data transmission and effectively utilize the bandwidth.

In an embodiment, the first message sending thread determines whether the first communication thread is transmitting data when receiving the information sending message, and if yes, placing the information sending message into the task pool, and if not, sending the information The message is passed to the first communication thread. When the first communication thread receives the data sent by the first target device, it is determined whether the first message receiving thread is processing information, and if so, the information receiving message is placed in the task pool, and if not, the information receiving message is delivered to The first message receives the thread.

For multi-threaded concurrent data reception and transmission for multiple target devices, the task pool POOL is used for caching and scheduling. The task pool is a global task pool and is used to cache information sending messages and information receiving messages corresponding to all target devices. The task pool can be an array linked list structure. Through the key operation lock operation, the thread security under multi-threading can be realized, and the dirty data can be prevented from being read and written. The receiving and sending operations that cannot be immediately performed can be put into the POOL for cache processing.

The global task processing thread polling the task pool is started for all the target devices. When the global task processing thread determines that the information sending message corresponding to one target device is cached in the task pool, it is determined whether the communication thread corresponding to the target device is idle, if If yes, the information sending message corresponding to the target device is taken out from the task pool, and sent to the communication thread corresponding to the target device.

When the global task processing thread determines that the information receiving message corresponding to a target device is cached in the task pool, determining whether the message receiving thread corresponding to the target device is idle, and if yes, extracting information corresponding to the target device from the task pool. The message is received and sent to the message receiving thread corresponding to the target device. Global task processing thread The information of the message receiving thread and the communication thread can be obtained by parsing the message receiving message and the message header of the information sending message.

The following uses the task pool to send a message and a message receiving message to the first target device as an example. 2 is a flow chart of processing an information transmission message according to an embodiment of the communication method of the present application, as shown in FIG. 2:

Step 201: The first message sending thread receives an information sending message for the first target device.

Step 202: Determine whether the first communication thread is transmitting data. If yes, execute step 203. If no, execute step 204.

Step 203: The first message sending thread puts the information sending message into the task pool. Go to step S205.

Step 204: The first message sending thread sends an information sending message to the first communication thread corresponding to the first target device. Go to step S207.

Step 205: The global task processing thread determines whether the first communication thread transmission data (or control command) ends, that is, whether the first communication thread is in an idle state. If yes, step 206 is performed, and if no, step 205 is continued.

Step 206: The global task processing thread extracts the information sending message from the task pool and sends the message to the first communication thread.

Step 207: The first communication thread sends data to the first target device according to the information sending message.

FIG. 3 is a flowchart of processing an information receiving message according to an embodiment of a communication method of the present application, as shown in FIG. 3:

Step 301: The first communication thread receives the data sent by the first target device, and generates an information receiving message.

In step 302, it is determined whether the first message receiving thread is processing data (or a control command). If yes, step 303 is performed, and if no, step 304 is performed.

Step 303: The first communication thread puts the information receiving message into the task pool. Go to step S305.

Step 304: The first communication thread sends an information receiving message to the first message receiving thread corresponding to the first target device. Go to step S307.

Step 305: The global task processing thread determines whether the first message receiving thread processes the data end, that is, whether the first message receiving thread is in an idle state. If yes, step 306 is performed, and if no, step 305 is continued.

Step 306: The global task processing thread retrieves the information receiving message from the task pool and sends the message to the first message receiving thread. Go to step S307.

Step 307: The first message receiving thread processes the information receiving message.

A flag bit can be set for each information sending message and information receiving message, which is used to identify the status and result of the operation of sending and receiving data, and can be set according to specific needs. For example, setting an identifier bit corresponding to the information transmission message and the information reception message, the identification bit identifies the processing status of the information transmission message and the information reception message, for example: 0-initial, 1-successful, -1-failed.

The message sending thread and the message receiving thread assign values to the flag according to the received feedback information, and adopt a locking mechanism. Each flag bit is transferred from one state to another state, and only one thread may be doing, and another thread may not exist. Move the same state to a different state. It can provide the flag bit query, and the user can obtain the result of the operation without calling the function again. It can reduce the waste of resources and time when the function call returns, and can also set the timing of the asynchronous transfer.

For example, the first message sending thread receives the feedback information corresponding to the sent data sent by the first communication thread, and assigns a status flag to the information sending message according to the feedback information. The first message receiving thread assigns a value to the status flag of the information receiving message according to the processing result of the information receiving message. The result of sending and receiving information can be determined by the flag bit, and the waiting flag is sent to the next transmission when the transmission is successful.

The global task processing thread extracts the information sending message and the information receiving message corresponding to the first target device from the task pool, and sends the message to the first communication thread and the first message receiving thread for processing. Determining whether the status flag of the information sending message and/or the information receiving message corresponding to the first target device taken from the task pool is successful, and if yes, the global task processing thread enters the task pool based on the information sending message and the information receiving message. In the sequence, the next information sending message corresponding to the first target device is taken out from the task pool, and the information receiving message is sent to the first communication thread and the first message receiving thread for processing. If not, the processing abnormality is prompted.

In an embodiment, the information sent by the first target device is received, the information includes a check code and a control command, and the control command sent by the first target device is stored in the memory block according to the received storage control command. In the process of storing the control command, the new check code corresponding to the control command is determined according to the control command receiving progress.

The new check code is compared with the received check code. If the check is successful, the check is successful, the information reception of the first target device is determined, the information transmission of the first target device is completed, and the control command is started. After determining to complete the information reception on the first target device, the received check code may be invalidated by the data, such as filling in 0, setting a random number, and the like. The new check code for determining the control command is the same as the algorithm for generating the check code by the first target device. For example, the new check code of the control command may be calculated by using the redundancy check code CRC32 algorithm.

By verifying the control command according to the check code in the received information, determining whether the information transmission is completed, and not being limited to the implementation logic of the underlying library, the processing flow can be reduced, thereby improving the operation efficiency and speed; Verification, using this type of verification when receiving information, can be more stable and faster.

In one embodiment, when the queried operation exception event is polled through the IB interface, it is determined whether an exception handling function corresponding to the exception event is registered. For example, it is judged whether or not a pointer of the exception handling function instance can be acquired, and if so, it is determined that the exception handling function corresponding to the exception event is registered, and if not, it is determined that the exception handling function corresponding to the exception event is not registered. If it is determined that an exception handler corresponding to the exception event is registered, the exception handler is automatically called back, and the exception handler handles the exception.

Exception handling is handled exceptions regardless of whether an exception handler corresponding to an exception event is registered. For example, when a transfer exception occurs, the transfer exception includes not only various exceptions that occur when the IB interface function is called, but also exceptions to the dispatch thread, deadlocks in the task pool, and so on. Corresponding processing for various transmission exceptions, including: command data retransmission, link termination and disconnection, thread stop and cleanup, etc. For example, when a serious transmission error occurs, the communication thread is stopped and disconnected, and when the connection is disconnected, when there is a new transmission task, the connection is reestablished, and the communication thread is started.

Callback function functions can also be provided for events. A callback function is a function called by a function pointer. For most events, there are basic processing functions. If the registered callback function is not empty, the callback function of the user-defined registration will be called back when the corresponding event occurs, so that the user can conveniently pay attention to the event of interest. Add the appropriate action. For example, setting a callback function for receiving a control command, when determining that the event is a receiving control command event, calling a corresponding callback function for receiving the control command, and directly processing the received control command event.

In one embodiment, multiple implementations can be devised for the communication method of the present application. As shown in FIG. 4, a multi-layer structure design can be adopted, the COMMON layer is encapsulated on the bottom layer of the IB VERBS interface function layer, and the SIMA main communication layer is encapsulated on the COMMON layer. The user can select the SIMA main communication layer or use the COMMON layer to do some customization of the operation functions as needed. The COMMON layer encapsulates the IB VERBS interface functions and provides some basic event processing logic and exception handling logic. The RMON operation can be performed directly using the COMMON layer.

The callback function is provided in the COMMON layer. The COMMON layer provides the corresponding operation error function pointer storage location, which can be used to store the function pointer for execution when an IB operation error occurs. The pointer is used by the SIMA master communication when the SIMA main communication layer is present. The error callback handler is registered (ie, points to this function). The COMMON layer obtains the location of the user-defined function pointer while providing the basic processing function according to the result of the IB VERBS interface function for sending and receiving data. When the COMMON layer detects that the location is not empty, the COMMON layer detects that the location is not empty. In the event of these events, these user-defined functions are called first, and then their basic processing is performed.

The SIMA main communication layer can be compatible with the startup mode of the MPI (Message Passing Interface) and can run under the mainstream platform and cluster management software. The SIMA main communication layer does not perform the retry operation when it fails to transmit, but processes it through the callback function or directly feeds back to the caller, and the caller decides whether to resend it. Both the sending data and the receiving data are performed asynchronously without blocking the call.

The communication method provided by the foregoing embodiment directly uses the IB VERBS interface or driver, and uses a multi-thread asynchronous operation for data transmission and reception of the target device, and can support an unequal number of receiving and transmitting data operations, thereby improving the speed of information transmission. , effectively use the bandwidth.

In one embodiment, as shown in FIG. 5, the present application provides a communication system 50, including: a thread configuration module 51, a data communication module 52, a processing status setting module 53, and an information verification module 54.

The thread configuration module 51 respectively creates a corresponding thread for at least one target device of the plurality of target devices, and the created thread corresponding to the target device includes a communication thread and a message processing thread, and the message processing thread includes a message sending thread and/or a message receiving thread. . The data communication module 52 communicates with the corresponding target device based on the created corresponding thread.

The first message sending thread sends an information sending message to the first communication thread, and the first communication thread sends the information to the first target device based on the information sending message by calling the IB interface. The first communication thread receives the information sent by the first target device by calling the IB interface, generates an information receiving message corresponding to the received information, and sends the information to the first message receiving thread. The first target device is a target device among the plurality of target devices, and the first communication thread, the first message sending thread, and the first message receiving thread are respectively a communication thread corresponding to the first target device, a message sending thread, and a message receiving thread. .

In response to determining that the first communication thread is currently in a busy state, the first message sending thread places an information transmission message to be sent to the first communication thread into the task pool. In response to determining that the first message receiving thread is currently in a busy state, the first communication thread places an information receiving message to be sent to the first message receiving thread into the task pool. The task pool is used to cache information transmission messages and/or information reception messages corresponding to multiple target devices.

The thread configuration module 51 creates a global task processing thread corresponding to a plurality of target devices. In response to determining that the first communication thread is currently in an idle state, the global task processing thread fetches the informational send message from the task pool and sends it to the first communication thread. In response to determining that the first message receiving thread is currently in an idle state, the global task processing thread fetches the information receiving message from the task pool and sends it to the first message receiving thread.

The processing status setting module 53 sets a status flag corresponding to the information transmission message, and sets a status flag corresponding to the information reception message. The first message sending thread receives the feedback information corresponding to the sending information sent by the first communication thread, and assigns a status flag to the information sending message according to the feedback information. The first message receiving thread assigns a value to the status flag of the information receiving message according to the processing result of the information receiving message.

The global task processing thread enters the task pool based on the information sending message and/or the information receiving message in response to the success of the status sending bit of the information sending message and/or the information receiving message corresponding to the first target device taken from the task pool. In sequence, the next information sending message and/or information receiving message corresponding to the first target device is taken out from the task pool and sent to the first communication thread and/or the first message receiving thread.

The information verification module 54 determines the new check code according to the received information (including the check code and the control command, etc.) according to the control command receiving the memory block for storing the control command, and the new check code and the receiving The check code to be verified is compared, and when the verification is successful, it is determined that the information reception of the first target device is completed. The information verification module 54 performs data invalidation processing on the received verification code in response to determining to complete the information reception of the first target device.

The thread configuration module 51 creates a first message sending thread in response to determining that the current information is sent as the first information transmission to the first target device. A first communication thread is created in response to determining that the first communication thread corresponding to the first target device is not currently created. In response to determining that the current information is received for the first time to receive information transmitted by the first target device, creating a A message receiving thread. Upon receiving an operation exception event through the IB interface, the first communication thread responds to determining that an exception handling function corresponding to the exception event is registered, and the exception handling function is called back.

The embodiment of the present application further provides an electronic device, such as a mobile terminal, a personal computer (PC), a tablet computer, a server, and the like. Referring to FIG. 6, there is shown a schematic structural diagram of an electronic device 600 suitable for implementing a terminal device or a server of an embodiment of the present application. As shown in FIG. 6, the computer system 600 includes one or more processors and a communication unit. Etc., the one or more processors, for example: one or more central processing units (CPUs) 601, and/or one or more image processing units (GPUs) 613, etc., the processors may be stored in a read-only memory ( Executable instructions in ROM) 602 or executable instructions loaded from random access memory (RAM) 603 from storage portion 608 perform various appropriate actions and processes. The communication unit 612 can include, but is not limited to, a network card, which can include, but is not limited to, an IB (Infiniband) network card.

The processor can communicate with the read-only memory 602 and/or the random access memory 603 to execute executable instructions, connect to the communication unit 612 via the bus 604, and communicate with other target devices via the communication unit 612, thereby completing the embodiments of the present application. An operation corresponding to any one of the communication methods, for example, an instruction for creating a corresponding thread for each of the plurality of target devices, and the created thread corresponding to any target device includes a communication thread and a message processing thread. A message processing thread includes a message sending thread and/or a message receiving thread; an instruction based on the configured corresponding thread to communicate with the corresponding target device. The communication process with the first target device includes: the first message sending thread sends an information sending message to the first communication thread, and the first communication thread sends the data to the first target device based on the information sending message by calling the IB interface; and/or, The communication thread receives the data sent by the first target device by calling the IB interface, generates an information receiving message corresponding to the received data, and sends the message to the first message receiving thread; the first target device is any one of the plurality of target devices. The first communication thread, the first message sending thread, and the first message receiving thread are respectively a communication thread corresponding to the first target device, a message sending thread, and a message receiving thread.

Further, in the RAM 603, various programs and data required for the operation of the device can be stored. The CPU 601, the ROM 602, and the RAM 603 are connected to each other through a bus 604. In the case of RAM 603, ROM 602 is an optional module. The RAM 603 stores executable instructions or writes executable instructions to the ROM 602 at runtime, the executable instructions causing the processor 601 to perform operations corresponding to the above-described communication methods. An input/output (I/O) interface 605 is also coupled to bus 604. The communication unit 612 may be integrated or may be provided with a plurality of sub-modules (e.g., a plurality of IB network cards) and on the bus link.

The following components are connected to the I/O interface 605: an input portion 606 including a keyboard, a mouse, etc.; an output portion 607 including, for example, a cathode ray tube (CRT), a liquid crystal display (LCD), and the like, and a storage portion 608 including a hard disk or the like. And a communication portion 609 including a network interface card such as a LAN card, a modem, or the like. The communication section 609 performs communication processing via a network such as the Internet. Driver 610 is also coupled to I/O interface 605 as needed. A removable medium 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory or the like, is mounted on the drive 610 as needed so that a computer program read therefrom is installed into the storage portion 608 as needed.

It should be noted that the architecture shown in FIG. 6 is only an optional implementation manner. In a specific implementation process, the number and type of components in FIG. 6 may be selected, deleted, added, or replaced according to actual needs; Different function components can also be implemented in separate settings or integrated settings, such as GPU and CPU detachable settings or GPU can be integrated on the CPU, the communication part can be separated, or integrated on the CPU or GPU. and many more. These alternative embodiments are all within the scope of protection disclosed herein.

In particular, the processes described above with reference to the flowcharts may be implemented as a computer software program in accordance with an embodiment of the present disclosure. For example, an embodiment of the present disclosure includes a computer program product comprising a computer program tangibly embodied on a machine readable medium, the computer program comprising program code for executing the method illustrated in the flowchart, the program code comprising Executing an instruction corresponding to the communication method step provided by the embodiment of the present application, for example, receiving an instruction for sending an information of a target device; transmitting an information sending message to an instruction of a sending thread corresponding to the target device; and sending a message according to the information sending message The underlying communication thread corresponding to the target device transmits a send data instruction, and the underlying communication thread sends data to the target device by calling the IB interface, and delivers the feedback information to the sending thread. In such an embodiment, the computer program can be downloaded and installed from the network via the communication portion 612, and/or installed from the removable medium 611. When the computer program is executed by the central processing unit (CPU) 601, the above-described functions defined in the method of the present application are performed.

The embodiment of the present application further provides a computer cluster. As shown in FIG. 7, the computer cluster includes: an IB switch 71 and a plurality of electronic devices 72, 73, 74, ... 75, 76, 77 as above. The IB switch 71 and the plurality of electronic devices 72, 73, 74, ..., 75, 76, 77 can be connected by a bus, a network cable, etc., each of which is provided with a communication portion (for example, an IB network card), and communication of each electronic device The department communicates with other electronic devices through the IB switch 71.

The communication method and system, the electronic device and the computer cluster provided by the foregoing embodiments adopt a multi-threaded manner for data transmission and reception of the target device, and receive and transmit data in an asynchronous manner, and provide an identifier bit to identify the result or state of the operation. Supporting the number of unequal reception and transmission data operations, can improve the speed of data transmission, and effectively use the bandwidth; adopt hierarchical structure design, provide RID as a data transmission means by calling IB interface or driver, provide custom callback support, convenient for users Customize operations and perform special functions to make communication and calculations more efficient.

The methods, apparatus, and apparatus of the present application may be implemented in a number of ways. For example, the methods, apparatus, and apparatus of the present application can be implemented in software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described sequence of steps for the method is for illustrative purposes only, and the steps of the method of the present application are not limited to the order specifically described above unless otherwise specifically stated. Moreover, in some embodiments, the present application can also be implemented as a program recorded in a recording medium, the programs including machine readable instructions for implementing the method according to the present application. Thus, the present application also covers a recording medium storing a program for executing the method according to the present application.

The description of the application is given for the sake of example and description, and is not exhaustive or limited to the present application. In the form disclosed. Many modifications and variations will be apparent to those skilled in the art. The embodiments were chosen and described in order to best explain the principles and embodiments of the embodiments of the invention,

Claims

A communication method, comprising:

Creating a corresponding thread for at least one target device of the plurality of target devices, the created thread corresponding to the target device includes a communication thread and a message processing thread, and the message processing thread includes a message sending thread and/or a message receiving thread;

And the corresponding target device communicates with the corresponding target device, wherein the communication process with the first target device includes: the first message sending thread sends an information sending message to the first communication thread, where the first communication thread calls the infinite bandwidth IB The interface sends information to the first target device based on the information sending message; and/or, the first communication thread receives the information sent by the first target device by calling an IB interface, and generates an information receiving message corresponding to the received information. And sent to the first message receiving thread;

The first target device is one of the plurality of target devices, and the first communication thread, the first message sending thread, and the first message receiving thread are respectively the first target device Corresponding communication thread, message sending thread and message receiving thread.
The method of claim 1, wherein the communication process with the first target device further comprises:

In response to determining that the first communication thread is currently in a busy state, the first messaging thread places the information transmission message to be sent to the first communication thread into a task pool; and/or,

In response to determining that the first message receiving thread is currently in a busy state, the first communication thread puts the information receiving message to be sent to the first message receiving thread into the task pool;

The task pool is configured to cache a to-be-processed information sending message and/or an information receiving message corresponding to the multiple target devices.
The method of claim 2, further comprising:

In response to determining that the first communication thread is currently in an idle state, a pre-created global task processing thread corresponding to the plurality of target devices retrieves the information transmission corresponding to the first target device from the task pool Sending a message to the first communication thread; and/or,

In response to determining that the first message receiving thread is currently in an idle state, a pre-created global task processing thread corresponding to the plurality of target devices retrieves the information corresponding to the first target device from the task pool A message is received and sent to the first message receiving thread.
The method of claim 3, wherein the communicating with the first target device further comprises:

Setting a status identifier bit corresponding to the information sending message;

The first message sending thread receives the feedback information corresponding to the sending information sent by the first communication thread, and assigns a status flag to the information sending message according to the feedback information; and/or,

Setting a status identifier bit corresponding to the information receiving message;

The first message receiving thread assigns a value to a status identifier of the information receiving message according to a processing result of the information receiving message.
The method of claim 4, wherein the communicating with the first target device further comprises:

Responding to the success of the information sending message corresponding to the first target device and/or the status flag of the information receiving message taken from the task pool, the global task processing thread is sent based on the information Sending a message and/or the information receiving message into the task pool, and extracting, from the task pool, the next information sending message corresponding to the first target device and/or sending the information receiving message Giving the first communication thread and/or the first message receiving thread.
The method according to any one of claims 1 to 5, wherein the communication process with the first target device further comprises:

Responding to the received information including the check code and the control command, determining a new check code according to a control command receiving progress of the memory block for storing the control command, and performing the new check code and the received check code The comparison is verified, and when the verification is successful, it is determined that the information reception of the first target device is completed.
The method according to claim 6, wherein the communication process with the first target device further comprises:

In response to determining to complete the information reception of the first target device, the received check code is subjected to data invalidation processing.
The method according to any one of claims 1 to 7, wherein the creating a corresponding thread for the first target device comprises:

Creating the first message sending thread in response to determining that the current information is sent as the first information transmission to the first target device; and/or,

Creating the first communication thread in response to determining that the first communication thread corresponding to the first target device is not currently created; and/or,

The first message receiving thread is created in response to determining that the current information is received for the first time to receive information transmitted by the first target device.
The method of any of claims 1 to 8, further comprising:

Upon receiving an operation exception event through the IB interface, the first communication thread calls back the exception handling function in response to determining that an exception handling function corresponding to the exception event is registered.
A method according to any one of claims 1 to 9, wherein:

The IB interface includes: an IB VERBS interface.
A communication system, comprising:

a thread configuration module, configured to create a corresponding thread for at least one target device of the plurality of target devices, where the created thread corresponding to the target device includes a communication thread and a message processing thread, and the message processing thread includes a message sending thread and/or a message receiving Thread

a data communication module, configured to communicate with a corresponding target device based on the created corresponding thread;

The first message sending thread sends an information sending message to the first communication thread, where the first communication thread sends information to the first target device based on the information sending message by calling an IB interface; and/or, the first communication The thread receives the information sent by the first target device by calling the IB interface, generates an information receiving message corresponding to the received information, and sends the information to the first message receiving thread;

The first target device is one of the plurality of target devices, and the first communication thread, the first message sending thread, and the first message receiving thread are respectively the first target device Corresponding communication thread, message sending thread and message receiving thread.
The system of claim 11 wherein:

In response to determining that the first communication thread is currently in a busy state, the first messaging thread places the information transmission message to be sent to the first communication thread into a task pool; and/or,

In response to determining that the first message receiving thread is currently in a busy state, the first communication thread puts the information receiving message to be sent to the first message receiving thread into the task pool;

The task pool is configured to cache a to-be-processed information sending message and/or an information receiving message corresponding to the multiple target devices.
The system of claim 12 wherein:

The thread configuration module is further configured to create a global task processing thread corresponding to the plurality of target devices;

The global task processing thread extracts the information sending message corresponding to the first target device from the task pool and sends the information to the first in response to determining that the first communication thread is currently in an idle state. a communication thread; and/or, in response to determining that the first message receiving thread is currently in an idle state, the global task processing thread fetches the information receiving message corresponding to the first target device from the task pool and Sended to the first message receiving thread.
The system of claim 13 further comprising:

a processing status setting module, configured to set a status identifier bit corresponding to the information sending message, and/or to set a status identifier bit corresponding to the information receiving message;

The first message sending thread receives the feedback information corresponding to the sending information sent by the first communication thread, and assigns a status flag to the information sending message according to the feedback information; and/or, the The first message receiving thread assigns a value to the status flag of the information receiving message according to the processing result of the information receiving message.
The system of claim 14 wherein:

Responding to the success of the information sending message corresponding to the first target device and/or the status flag of the information receiving message taken from the task pool, the global task processing thread is sent based on the information Sending a message and/or the information receiving message into the task pool, and extracting, from the task pool, the next information sending message corresponding to the first target device and/or sending the information receiving message Giving the first communication thread and/or the first message receiving thread.
A system according to any one of claims 11 to 15 further comprising:

The information verification module is configured to: in response to the received information including a check code and a control command, determine a new check code according to a control command receiving progress of a memory block for storing the control command, and generate a new check code and receive The check code to be checked for comparison, and when the verification is successful, it is determined that the information reception of the first target device is completed.
The method of claim 16 wherein:

The information verification module is further configured to perform data invalidation processing on the received check code in response to determining to complete information reception on the first target device.
A system according to any one of claims 11 to 17, wherein

The thread configuration module is further configured to create the first message sending thread in response to determining that the current information is sent as the first information transmission to the first target device; and/or, in response to determining that the current information is not created and Creating, by the first communication thread corresponding to the first target device, the first communication thread; and/or, in response to determining that the current information is received for the first time to receive information sent by the first target device, creating the first message Receive thread.
A system according to any one of claims 11 to 18, wherein

Upon receiving an operation exception event through the IB interface, the first communication thread calls back the exception handling function in response to determining that an exception handling function corresponding to the exception event is registered.
A system according to any one of claims 11 to 19, wherein:

The IB interface includes: an IB VERBS interface.
An electronic device, comprising: a processor, a memory, an IB communication unit, and a communication bus, wherein the processor, the memory, and the communication unit complete communication with each other through the communication bus;

The memory is for storing at least one executable instruction that causes the processor to perform operations corresponding to the communication method of any of claims 1-10.
The electronic device of claim 21, wherein the IB communication portion comprises an IB network card, and the processor communicates with other external electronic devices via the communication bus and through the IB network card.
A computer cluster, comprising: a plurality of electronic devices according to claim 21 or 22, and switching devices respectively connected to the respective electronic devices, each of the electronic devices passing through the respective IB communication portion and being exchanged The device communicates with other electronic devices.
A computer program comprising computer instructions that, when executed in a processor of a device, perform the steps of the communication method of any of claims 1-10.